Automate Before You Delegate to AI

AI is not automation

There's a conflation happening in a lot of businesses right now. "AI" and "automation" are being
used interchangeably, and it's causing teams to reach for complex, expensive solutions to
problems that have simple, cheap ones.

Automation moves data and triggers actions based on rules. AI makes judgements. These are
different tools for different problems, and mixing them up wastes money and adds fragility.

The test

Before you bring in an LLM, answer this: is the decision you're trying to automate based on
rules, or does it require judgement?

If a new lead fills out your form and you want to:
- Add them to your CRM → rule
- Send them a welcome email → rule
- Assign them to a sales rep based on company size → rule
- Notify your team in Slack → rule

None of that requires AI. All of it can be done with n8n, Zapier, or Make in an afternoon.

If you want to:
- Read an unstructured email and extract intent → probably AI
- Score a lead based on a nuanced set of signals → maybe AI
- Draft a personalised response based on context → AI

The distinction is whether the logic can be written as if/then statements. If it can, you don't
need AI.

The cost difference is significant

A well-built n8n workflow running on your own server costs roughly $20/month to operate at
significant scale. An equivalent workflow routing everything through GPT-4 might cost 50x that,
runs slower, and fails in ways that are harder to debug.

Automation errors are usually obvious: a field didn't map, a webhook didn't fire, a condition
didn't match. AI errors are subtle: the model misclassified something, hallucinated a field
value, or behaved differently because the prompt drifted.

Simpler tools fail louder. That's a feature.

Where automation ends and AI begins

There's a natural handoff point. Automate the deterministic parts of your workflow — data
movement, notifications, routing, transformations. Then use AI only at the decision points that
genuinely require reading and reasoning over unstructured content.

A practical example: a support ticket comes in. Automation routes it to the right queue based on
the subject line keyword. AI reads the body, classifies the severity, and drafts a suggested
reply. Automation sends the reply once a human approves it.

The AI is doing one thing it's actually good at. Everything else is rules.

Start with a process map

Before you automate anything, draw the process. Every step. Every handoff. Every decision point.

You'll usually find that 80% of the steps are pure data movement — no intelligence required.
Automate those first, get them working reliably, and then evaluate whether the remaining 20%
actually need AI or just better-structured rules.

The teams that build the most reliable automated systems aren't the ones using the most AI.
They're the ones who are ruthlessly honest about what actually requires it.